In transmission or communication system, signals originating from a far end of a communication circuit are received at a near end of the circuit. The received signals, either electrically or acoustically, may find their way into the return path along with near-end input. Far-end reception of the near-end input may include an attenuated and delayed reflection, i.e., an echo, of the original far-end input signals. In telephone systems, whether wired or wireless, these echo phenomena can make a conversation unintelligible. In data communication systems, again whether wired or wireless, such echoes can cause errors in transmission or otherwise degrade throughput performance.
Adaptive filters are used in numerous applications to remove undesired frequency content from a signal and are used in telecommunication systems as echo cancellation systems to remove from a signal echoes that may arise as a result of the reflection and/or unwanted coupling of input signals back to the originator of the input signals. For example, echoes occur in instances where signals that were emitted from a loudspeaker are then received and retransmitted through a microphone, i.e., acoustic echo, or when reflections of a far-end signal are generated in the course of transmission along wiring junctions where impedance mismatch occur, i.e., line echo.
Presently, an adaptive finite-impulse response (FIR) filter may be used to reduce or eliminate the echo where the echo cancellation characteristics are defined in the International Telecommunication Union-Standardization Sector (ITU-T) Recommendations G.165 and G.168 and the contents of each of the foregoing ITU Recommendations being incorporated herein by reference as if set forth in full. FIG. 1 illustrates a functional block diagram of an echo cancellation circuit 150 interposed between the far end and the near end of a telecommunication system 100 where the echo cancellation circuit operates at a near end 102. The near-end input 130 to return signal 132 is shown as combining with a near-end echo signal 122 represented by the far-end input signal 110 as filtered by, that is, as attenuated and delayed by, the z-domain echo transfer function 120, H(z). The resulting return signal 132 is represented then as a linear combination of the near-end input signal 130 and the near-end echo signal 122.
Accordingly, when a digital representation of the echo transfer function is in the form of an adaptive FIR 156, and the gains are adjusted to mimic the echo transfer function 120, when the far-end input signal 110 is filtered by the adaptive FIR 156, the resulting signal 158 may be differenced with the return signal 132 to cancel the echo from the return line signal 132. As illustrated in FIG. 1, to accomplish this canceling effect, the post-cancellation return line signal 140 is directed into a nonlinear processing module 152 that may pre-filter background or ambient noise and establish a threshold above which little or no adaptation of the IIR filter is permitted. The threshold logic is us used to address the double talk situations where the return signal may have both near-end input and echo in temporal proximity. If the pre-filtered return line signal is below the threshold, it may be used, at each step k, as an error signal 154, ek, to drive the adaptation of the gains the FIR filter 156. The FIR filter may require several delay states with each output or input state being tapped, multiplied by a gain and summed. The gains for the FIR filter may be chosen to represent the most likely echo transfer function 120, H(z), and subsequently, these gains may be adjusted by relationships driven by the error signal, ek. Least-mean-square (LMS) adaptive algorithms are commonly implemented in adaptive cancellation devices to adjust the gains of the adaptive FIR filter. A FIR filter may be represented as
                                          H            k                    ⁡                      (            z            )                          =                              ∑                          n              =              0                        L                    ⁢                                    h                              n                ,                k                                      ·                          z                              -                n                                                                        [        1        ]            
A typical rule of adaptation or adjustment of the FIR filter gains is to use the product of the error signal 154, the normalized input signal, and a step size, or adapting gain, β, to adjust the gains. For example, for each filter coefficient, n, where n=0, 1, 2, . . . L:hn,k+1=hn,k+β*e(k)*xk−1/xmax.  [2]
FIR filters typically require a long tap delay to model effectively an echo return path. FIR filters, while stable representations of all zero transfer functions, are typically slow to adapt, require more memory than recursive filters memory, and, due to the number of taps, can be computationally cumbersome.
With certain types of input signal, such as human speech, are characterized by the dominance of distinct peaks followed by a long decay over time. A majority of the computation is devoted to FIR coefficient update on the long decay portion of the signal, which actually contributed little significance to the actual echo energy. In addition, performing aggressive adaptive filtering on these low energy decays actually causes error in estimation in many types of adaptive FIR filters, e.g. normalized LMS filters, and degrades the overall echo cancellation performance.
Infinite impulse response (IIR) filters, or recursive filters, are implemented forms of pole-zero transfer functions that do not require a long tap delay. Typically, IIR filters are used to numerically mimic very specific echo return paths in which stability of the pole-zero transfer function can be guaranteed during adaptation. In addition, the poles must be properly represented numerically and thus practical embodiments in digital signal processing require a high degree of precision in implementation because small bit errors can cause large filter errors including instability. Methods of adaptation mechanisms are known to those of ordinary skill in the art and are found described in Adaptive Signal Processing, by Bernard Widrow and Samuel D. Stearns, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1985, particularly pages 99-101 and 154-161.
Accordingly, there remains a need for the rapid convergence of an IIR filter and the stability of an FIR filter to be applied to echo cancellation. The present invention, in its several embodiments provides echo cancellation using an adaptive IIR filter and an adaptive FIR filter.